wipke@secs.ucsc.edu (03/13/91)
Is the Discover Magazine article about Brooks' work accurate? Are there other views not properly characterized in this fine article? /sig
minsky@media-lab.MEDIA.MIT.EDU (Marvin Minsky) (03/14/91)
In article <13MAR91.12593277@secs.ucsc.edu> wipke@secs.ucsc.edu writes: >Is the Discover Magazine article about Brooks' work accurate? >Are there other views not properly characterized in this fine >article? >/sig The article is not very accurate.
loren@ingrid.llnl.gov (Loren Petrich) (03/15/91)
In article <5511@media-lab.MEDIA.MIT.EDU> minsky@media-lab.media.mit.edu (Marvin Minsky) writes: >In article <13MAR91.12593277@secs.ucsc.edu> wipke@secs.ucsc.edu writes: >>Is the Discover Magazine article about Brooks' work accurate? >>Are there other views not properly characterized in this fine >>article? >The article is not very accurate. Why is that? I know that the article contains some quotes from Minsky knocking Brooks's work as a step backwards toward insect intelligence. If Marvin Minsky himself feels that his position has been misrepresented, then he should explain why. His charge that a robot that can move around but cannot recognize a Coke bottle (?) is no good may only indicate additional work to be done -- such as to add an artificial-vision system. Many of Brooks's robots appear to be blind. I think that the difficulty that Brooks's critics have with his work is because it is not based on the sort of AI they would think appropriate -- a system that builds a mental model of the robot and the world around it and works out appropriate responses from that. In my mind, that precisely parallels the difficulties that certain critics have had with Neural Networks -- that NN's do not build a mental model of what they are learning. Yet however desirable that mental-model, symbol-processing AI may be, it still has a long way to go. How well can they "learn" from experience? My interest in NN's is derived from that fact that they _can_ do what might be called "learning", even if it is not on a terribly advanced level. $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ Loren Petrich, the Master Blaster: loren@sunlight.llnl.gov Since this nodename is not widely known, you may have to try: loren%sunlight.llnl.gov@star.stanford.edu
nagle@well.sf.ca.us (John Nagle) (03/17/91)
The great strength of Rod Brooks' group is that they are quite good
at building machines, or what the Japanese call "mechatronics". The
first legged machine was built by Brooks and one grad student in six
weeks. That's quite an achievement. The integration between the
electronics and the mechanics in that lab is quite good. There
are many little details handled well there. Using the M68HC11 SPI
lines as a token ring and bringing out the error signal from R/C
servos comes to mind.
Working small seems to be a win. Mobile robotics in the '80s
suffered badly from domination by the Army Tank Command's influence
via the Strategic Computing Initiative (or "Star Wars for the ground").
The field moved to big vehicles, including one group with a Bradley
Fighting Vehicle. Several groups had truck-sized machines, and one group
(at OSU) built a multiton four-legged walker. Work on this heavy metal
diverted effort from the artificial intelligence end of the problem,
as well as making testing an outdoor job.
Superior engineering skills and good engineering judgement
account for much of the success of Brooks' efforts.
John Nagle